Skip to content

Roadmap ​

Honest scope from the project's planning records — no invented dates, ever. This page tracks the same phase structure the maintainer works from; when planning changes, this page changes with it. Nothing on this page is a promise with a deadline; each item is a direction with an entry condition.

Release status: package 6.1.0 is the current stable line. 3.0.0, 4.0.0, 5.0.0, 6.0.0 and 6.0.1 remain published and immutable; a tag alone is not a new release, and republishing an existing version is not a path forward.

Where Mjölnir is now ​

  • v6.1.0 — current stable. It fixes reports, config validation and release workflows that misbehaved silently, and makes two config checks strict (ignore[].expires and plugins). 6.0.1 repaired the CI reporting 6.0.0 shipped broken. The 6.0 line: a first run reads GATE or WARN per finding and nothing else, and it names the command that turns the scan into a blocking check. The core / extended / quarantine tiers, the E0–E2 evidence rungs and the L0–L5 trust rungs stay in --json and mjolnir explain — they are the differentiator, not first-run comprehension. --strict is now --include-warn, because the old name promised enforcement the flag never had: a quarantined detector is advisory by design.79 rules, 73 of them carrying a false-positive rate measured against real OSS code (FP-AUDIT); the unmeasured remainder is quarantined, never silently shipped. mjolnir share writes one self-contained HTML trust report — no server, no network, byte-stable under the deterministic clock.
  • v5.1.0 — prepared, never published; 6.0.0 superseded it. The reporting surface is checked against the working tree rather than against a plan, and each check is a gate in npm run certify. A silent capability loss is disclosed through a degradation ledger rather than absorbed, and a verdict is never published without a measurement behind it. All of it shipped in 6.0.0. Package publication does not imply Trust certification: the release-readiness record is separate and still names what is open.
  • JSON report (schemaVersion: 1), exit codes and CLI surface are frozen contracts.
  • The scan core is deterministic and zero-network: the same input produces the same verdict, and nothing leaves the machine.

The trust-engineering train — shipped into the 3.x line ​

An eleven-increment engineering train hardened the product's release trust end-to-end. It is part of the 3.x line; the table below remains as the historical release train that led into the 3.0 cutover.

ReleaseContent
1.1.0Azure DevOps Pipelines as a first-class scan target — adapter, detection and CI-workflow rules
1.1.1Jenkins (root Jenkinsfile) as a first-class scan target
1.1.2Quarantine remediation: a generated, drift-locked ledger plus rework-or-retire of the worst measured rules (QUARANTINE-REMEDIATION)
1.1.3Machine-verified blast-radius audit and containment fixes (BLAST-RADIUS-AUDIT)
1.1.4The Trust Constitution and the mjolnir ci release-trust verdict gate (RELEASE-TRUST-CONTRACT)
1.1.5A false-green attack corpus: adversarial suites that prove the scanner cannot be fooled into a clean verdict
1.1.6Evidence-graph provenance binding, scope integrity, and exit-code mutation proofs
1.2.xPlaywright trace.zip forensics · the forensic verdict taxonomy and Selector Health v2 · the Playwright capability matrix as a product surface
1.3.0MCP runtime-evidence tools (forensics, triage, pw-report) and the agent safety contract
1.4.0Trust Artifact identity binding and the full HTML artifact
2.0.02.0 preparation: breaking-set inventory, migration draft, boundary-law guards

Next ​

One promise per version, in order. Each is a net subtraction of surface or vocabulary before it is an addition of feature, and each carries a kill criterion: if it cannot meet its exit gate, it is abandoned at that criterion rather than half-shipped on top of the version before it.

VersionThemeThe promiseThe number it moves
7.0The gate proves itselfThe PR output is the change in trust, not the state — what this branch introduced and removed, against the merge-base.Every gated PR proves it blocks; the demo repo blocks in one CI run
8.0One notch widerEvery shipped language and CI provider measured, or not shipped. A rule enters the default scan only if it can carry a measured FP rate.Zero new default-scan rules without n ≥ 10 measured FP verdicts
9.0It ranFindings reach L3–L5 from a real run report, read from disk, parsed and discarded. No database, no history store, no dashboard.Share of GATE findings reaching L3+, measured on fixtures
10.0Prove it on adversarial codeA public, versioned, licensed false-green benchmark: repositories where a green CI is provably wrong, each with its expected finding set.Every default-scan rule has a benchmark row with its measured FP rate
  • Publish only registry-backed claims: a git tag is not public install guidance until the matching npm version and GitHub Release are live. VERSIONING remains the contract for what "stable" means.

What this roadmap will never contain ​

  • Dates attached to unbuilt features.
  • A light theme (dark-only is a design decision, not an omission).
  • Telemetry that isn't opt-in.
  • Rules without fixtures, or a tier system that moves on anything but measured evidence.

Released under the MIT License.